Extension Models
Extensions can bring their own AI models into Raycast. Install an extension that provides models, allow it to run in the background, and its models join the model picker next to everything else in Raycast AI. It's the best way to reach inference from a service you already use, a model Raycast doesn't offer, or a setup you've built yourself.
Requests to extension models are handled by the extension, without passing through Raycast, and don't use your AI credits. Check the extension's Store page to see which service it connects to.
Get Started
Several extensions in the Store already provide models, including Ollama AI, AI by Vercel, Cerebras, Groq, LM Studio, and OpenCode. Extensions that provide models show up in the model picker as soon as they're installed, waiting for you to allow them:
- Install an extension that provides AI models from the Store.
- Open the model picker in AI Chat, Quick AI, or AI Commands, then search the extension you installed listed with an Allow AI Models label.
- Select it. Raycast asks whether the extension can run in the background to discover and refresh its models. Choose Allow Background Refresh.
- If the extension needs preferences, such as an API key, or asks you to sign in, Raycast walks you through that. Once setup finishes, the extension's models appear in the picker, ready to use.
If you're not ready yet, choose Not Now and the extension stays in the picker for later. Choose Don't allow to turn its models off. You can change your mind at any time in Settings.
Finishing Setup
An extension stays under Set Up AI Model Providers until it's ready to provide models. The label next to it tells you what's missing:
- Allow AI Models: You haven't decided whether the extension can provide models yet.
- Preferences Required: The extension needs preferences, such as an API key, before it can list its models.
- Finish Setup: You allowed the extension, but setup didn't complete, for example because you closed a sign-in window. Select it to try again.
Using Extension Models
Extension models work in Quick AI, AI Chat, and AI Commands. Pick them from the model picker like any other model, where they're grouped under the extension's name.
Each models' capabilities varies on the extension and model. Depending on what it declares, a model can accept a system prompt, image attachments, a temperature setting, and a reasoning effort, and it can use AI Extensions and MCP servers during a conversation. If a model doesn't support images, Raycast leaves any images out of the request. Web search and image generation aren't available to extension models.
Extension models aren't available to other extensions through the Extension AI API. Only the extension that provides a model can use it in its own commands.
Keeping Models Up to Date
Raycast keeps an extension's models up to date by running it in the background. A refresh happens when you install or update the extension or change its preferences, when Raycast updates its own list of models, and whenever the extension requests one. If the extension stops providing a model, it's removed from the picker.
If a refresh fails, for example because the service is unreachable, Raycast keeps the models it already had so you can carry on working.
Managing Extension Models
Disabling or uninstalling the extension removes its models from Raycast AI. You can turn an extension's models on or off from two places in Settings:
- Settings → AI → Models & Providers → Manage Models: Each extension that provides models has its own section with a switch next to its name, and you can disable individual models here.
- Settings → Extensions: Select the extension and use the Allow AI Models switch. The description below it shows when the models were last refreshed and how many there are, or the error if the last refresh failed. You can also manage the API key or sign in for the extension here too.
For Developers
Any extension can provide models by declaring a model provider in its manifest and implementing two functions: one that lists the available models and one that streams responses. Read Provide AI Models in the developer documentation to get started.